I've been running multiple AI agents in parallel to build software. One planned, one wrote code, one tested, one reviewed. More agents should mean more output.
It didn't.
Past a certain number, adding agents made things slower. They didn't fail. They waited for me to make a decision, answer a question or approve a direction. The agents could work around the clock. I couldn't make decisions that fast.
So I stopped asking how many agents I could run. I started asking how many decisions I could make well.
Leaders run into the same limit
Every first-time manager goes through this. You were fast as an engineer because you did the work yourself. Now your team does the work, and you've become the place where it piles up. A pull request waits for your review. A design waits for your nod. A small question sits in your DMs for two days.
Nobody is blocked by a lack of talent. They're blocked waiting on you.
The fix isn't more effort. It's fewer decisions only you can make.
Here's what helped me, with agents and with people:
- Make your intent clear up front. When agents had a clear spec, they asked me fewer questions. People work the same way. When the "why" and the boundaries are clear, people can make the call themselves.
- Separate one-way doors from two-way doors. Most decisions are easy to undo. Let the team make those. Keep your attention for the few that are hard to reverse.
- Make waiting visible. I set up notifications so I could see exactly where an agent needed me. Do the same with your team, and ask: what is waiting on me right now?
- Size the team to your attention. If your direct reports all need you every day, you don't have a team. You have a queue.
Culture is what happens when you're not in the room
A team that can only move when the leader decides has a bottleneck, however good the leader is. A team that understands the intent well enough to decide on its own can move quickly.
AI didn't create this problem. It made it obvious. When the work moves at machine speed, every handoff that waits for a person becomes visible.
The question isn't how much more you can take on. It's how much you can get out of the way of.
Enjoyed this article? Get new writing on engineering culture and the craft of leading teams, delivered to your inbox.
Subscribe →